Qubole Cross-Account IAM Credential Compromise Grants Attacker Access to Dozens of AWS Customer Environments
Between October 14 and 26, 2026, a threat actor used compromised IAM credentials from AWS account 805246085872, associated with Qubole's data platform, to assume cross-account IAM roles in dozens of downstream customer AWS environments [1][3][4]. Once inside each tenant, the attacker followed an identical sequence: enumerate resources, harvest secrets, validate stolen credentials with TruffleHog, and bulk-download S3 data [1][11]. Wiz published its findings on November 9, 2026, after Sentra initially detected suspicious activity across several of its own customer environments and shared the information with Wiz and Mandiant [1][5]. Sentra confirmed dozens of corporate victims and millions of files stolen from at least one organization [11]. AWS stated that its "services were not compromised by this issue and operated as designed" [6].
On November 11, 2026, an actor calling themselves "n0n" claimed responsibility for the intrusion [1][7][9]. The n0n group also claimed to have stolen 53.9 GB across 30,743 files directly from Qubole, including customer contracts, pricing records, staff and trial-user information, and production data, and alleged it holds nearly one petabyte of data from Qubole's customer networks [7]. Those volume claims remain unverified. On November 16, n0n named 59 organizations on its leak site, including JPMorgan Chase, Visa, and Mastercard, and began releasing data [8]. Check Point Exposure Management assesses with moderate confidence that the claims against JPMorgan Chase, Visa, and Mastercard trace back to one or more shared data vendors, not to direct intrusions into those companies [8]. The released data structure contains no Qubole-specific markers, so the link cannot be confirmed or ruled out [8].
Background: Qubole and Cross-Account IAM Trust
Qubole is a cloud data lake and analytics platform owned by Idera Inc. [6][7]. It helps enterprises manage large datasets and run machine learning workloads, with deep AWS integration for compute resource management [3]. Many SaaS platforms that operate inside a customer's AWS account, Qubole included, rely on cross-account IAM roles. During onboarding, the customer creates a role in their own AWS account with a trust policy that allows the vendor's AWS account to assume it, typically gated by an External ID [1][4]. When the vendor needs to act in the customer's environment, an identity in the vendor's account calls sts:AssumeRole with the role ARN and External ID and receives temporary credentials scoped to that role's permissions [1][3].
The critical detail in this incident: many of the abused roles were dormant. Customers had stopped using Qubole but never removed the trust relationship. Sentra noted that "Qubole roles were still in place long after companies had stopped using Qubole, and those roles could read nearly all of the data the attacker went after" [4][5]. The trust policies only checked things the attacker already possessed: the correct source account and the correct External ID [12]. Nothing in the trust policy distinguished the attacker from Qubole [12].
Threat Actor: n0n
n0n is a cyber extortion group active since September 2026 [8]. Catalin Cimpanu of the Risky Business newsletter reported that n0n claimed responsibility for the Qubole intrusion [1]. The group's leak site named 59 organizations as of November 16, 2026 [8]. The leaked samples share a common file structure across brands: campaign directories, engagement events, member profiles containing PII, member bookings, and program counts, pointing to a shared source system or vendor export [8]. The data appears to be marketing and loyalty records, not payment data [8].
JPMorgan Chase confirmed that third-party vendors were targeted and that the attackers may have accessed marketing data including some customer names and email addresses [7][9]. The leaked records appeared to contain about 16,000 customer email addresses tied to Chase's invitation-only Reserve credit card program, along with some phone numbers and Chase account IDs [9]. Visa stated the claims concerned a service provider in the Asia-Pacific region and that its systems were unaffected and no payment data was exposed [7]. Mastercard said a provider incident exposed a limited number of email addresses from cardholders enrolled in a benefit program [7]. None of these financial institutions confirmed Qubole as the specific vendor.
The hacking group claimed it contacted Qubole and some of the affected companies before the public reports from Sentra and Wiz, but their extortion demands were ignored [7]. As of November 11, 2026, Qubole and Idera had published no statement [6].
Technical Detail: Initial Access and Credential Compromise
Wiz has not disclosed how the attacker obtained credentials for the two IAM users in Qubole's AWS account [12]. What is confirmed is that the first suspicious sts:AssumeRole activity originated on October 14, 2026, from an IP address associated with Mullvad VPN, targeting customer integration roles from AWS account 805246085872 [1][2]. The requests used IAM user AIDA3W7DIFLYOLTK63SFS [1]. Wiz states that in environments it observed, only the threat actor used this IAM user, so any activity from it warrants investigation [1].
On October 18, the attacker began using a second IAM user, AIDAJYRRH5Y3A7AAJM2HW, which also carries legitimate Qubole platform traffic from AWS infrastructure [1]. This second identity was used alongside a different Mullvad VPN address [2].
Wiz also described separate activity from October 8 to October 11, in which a different actor used customer-owned AWS access keys believed to have been created for Qubole integrations [1][6]. The relationship between this earlier activity and the October 14-26 campaign is not confirmed.
All observed attacker activity originated from non-AWS infrastructure: Mullvad VPN exits and an OVH-hosted server [4][10]. None of it originated from Qubole's own network [4].
Technical Detail: Role Assumption and Enumeration
Every malicious sts:AssumeRole request included a valid External ID and succeeded [5][6][12]. The External ID is not a secret; AWS and Wiz note it is often visible in the role configuration, and it is logged in plaintext in CloudTrail [5][6]. An External ID alone does not constitute a strong access control. It prevents the "confused deputy" problem in legitimate integrations but provides no protection when the caller already holds valid IAM credentials from the trusted account.
The attacker used role session names following a consistent format: the prefix quiam-, followed by a suffix describing the action being taken. Session names observed include quiam-enum and quiam-scope-full [4]. This naming pattern is a reliable CloudTrail hunting indicator.
Wiz reports that in a majority of affected environments, the attacker's requests were the only AssumeRole activity from Qubole's account in October [12]. The roles had been dormant for months or longer, meaning baseline activity was zero and any call at all was anomalous.
Technical Detail: Secrets Harvesting and Credential Validation
Once inside each customer environment, the attacker followed a repeatable playbook [11]:
- Resource enumeration: Listed S3 buckets, databases, EC2 instances, security groups, and IAM identities [2][4].
- Secrets collection: Read secrets from AWS Secrets Manager, SSM Parameter Store, and Lambda environment variables [1][3][6]. Sentra counts about 100 passwords and keys read in 7 minutes in one observed session [5].
- Credential validation: Used TruffleHog to call
GetCallerIdentitywith discovered IAM user access keys, validating them in bursts lasting seconds [1]. Stolen credential types included AWS access keys, Microsoft Entra ID service principal credentials, and Okta OAuth client credentials [2][6]. - Data exfiltration: Downloaded files from accessible S3 buckets, in some cases entire buckets [4][10][13].
Sentra observed no changes to or deletion of customer data during the theft it documented [9]. The attacker appeared to focus on collection rather than disruption.
Technical Detail: Infrastructure and Anonymization
The attacker routed activity through Mullvad VPN, an OVH-hosted server, and a Datacamp-hosted server [4][10]. Four IP addresses are documented in the Wiz research:
45.83.220.211(Mullvad VPN, Sweden) [1]185.209.199.73(Mullvad VPN, Sweden) [1]57.129.54.120(OVH VPS, Germany) [1]89.37.63.175(Datacamp-hosted, Sweden) [1]
Indicators of Compromise
| Type | Value | Context | Source |
|---|---|---|---|
| AWS Account ID | 805246085872 |
Source account for all observed malicious role assumptions (Qubole platform account) | [1] |
| IAM Principal ID | AIDA3W7DIFLYOLTK63SFS |
First IAM user; seen only in threat actor activity in observed environments | [1] |
| IAM Principal ID | AIDAJYRRH5Y3A7AAJM2HW |
Second IAM user; active from October 18; also carries legitimate Qubole traffic | [1] |
| IP Address | 45.83.220.211 |
Mullvad VPN exit used for AssumeRole calls | [1] |
| IP Address | 57.129.54.120 |
OVH VPS used by threat actor | [1] |
| IP Address | 185.209.199.73 |
Mullvad VPN exit used to assume integration roles | [1] |
| IP Address | 89.37.63.175 |
Datacamp-hosted IP used to assume customer roles | [1] |
| Session Name Prefix | quiam- |
Role session name prefix used across all observed attacker sessions | [4][5] |
| Tool | TruffleHog | Open-source credential scanner used to validate harvested keys via GetCallerIdentity |
[1][2] |
MITRE ATT&CK Mapping
The following techniques are directly supported by the source material describing this campaign:
| Technique ID | Name | Application in This Campaign |
|---|---|---|
| T1199 | Trusted Relationship | Attacker used compromised Qubole credentials to exercise pre-existing cross-account IAM trust [1][3] |
| T1078.004 | Valid Accounts: Cloud Accounts | Two legitimate Qubole IAM users were used to authenticate and assume roles [1][2] |
| T1550.001 | Use Alternate Authentication Material: Application Access Token | Temporary STS credentials obtained via AssumeRole were used to operate within victim environments [1][3] |
| T1087.004 | Account Discovery: Cloud Account | Attacker enumerated IAM identities within victim environments [2][4] |
| T1526 | Cloud Service Discovery | Attacker listed S3 buckets, EC2 instances, databases, and security groups [2][4] |
| T1552 | Unsecured Credentials | Secrets retrieved from Lambda environment variables, which store credentials in runtime configuration rather than traditional files [1][6] |
| T1552.007 | Unsecured Credentials: Cloud Secrets Management Stores | Secrets harvested from AWS Secrets Manager and SSM Parameter Store [1][3][6] |
| T1530 | Data from Cloud Storage | S3 data downloaded, sometimes entire buckets [4][10][13] |
| T1090.003 | Proxy: Multi-hop Proxy | Activity routed through Mullvad VPN and OVH infrastructure [4][10] |
CloudTrail: AssumeRole from Qubole Account
The primary detection artifact is AssumeRole events in AWS CloudTrail where requestParameters.roleSessionName begins with quiam- or where the source account is 805246085872 [1][4][5]. Wiz recommends reviewing CloudTrail from October 1, 2026 in every account that holds a Qubole integration role [5].
A targeted CloudTrail query for SIEM platforms (Splunk syntax):
index=cloudtrail eventName=AssumeRole
| where like(requestParameters, "%805246085872%") OR like(requestParameters, "%quiam-%")
| table _time, recipientAccountId, requestParameters, sourceIPAddress, userIdentity.arn
| sort - _time
For environments using Amazon Athena against CloudTrail logs stored in S3:
SELECT eventtime, recipientaccountid, requestparameters, sourceipaddress, useridentity.arn
FROM cloudtrail_logs
WHERE eventname = 'AssumeRole'
AND (requestparameters LIKE '%805246085872%'
OR requestparameters LIKE '%quiam-%')
ORDER BY eventtime DESC;
Detecting Post-Assumption Enumeration
The attacker followed a consistent sequence after assuming each role. Look for the following API calls occurring within minutes of a cross-account AssumeRole event:
ListBuckets,ListObjects(S3 enumeration)DescribeInstances,DescribeSecurityGroups(EC2 enumeration)ListRoles,ListUsers,GetAccountAuthorizationDetails(IAM enumeration)GetSecretValue(Secrets Manager harvest)GetParameter,GetParameters,GetParametersByPath(SSM Parameter Store)GetFunctionor LambdaGetFunctionConfiguration(Lambda environment variable access)
Sentra documented approximately 100 passwords and keys read in 7 minutes in one session [5]. A burst of GetSecretValue or GetParameter calls across multiple secret names within a short window, particularly following a cross-account role assumption, is a strong behavioral indicator.
Credential Validation via GetCallerIdentity
The attacker used TruffleHog to call sts:GetCallerIdentity with discovered IAM access keys in rapid bursts [1]. This call is benign on its own but, when issued in rapid succession against multiple distinct access key IDs, indicates programmatic credential validation. Alert on:
- Multiple
GetCallerIdentitycalls from a single source IP within seconds GetCallerIdentitycalls where the access key ID in the request does not match the caller's own identity
Network-Layer Indicators
Search proxy, VPN, and firewall logs for connections to or from the four documented IPs: 45.83.220.211, 185.209.199.73, 57.129.54.120, and 89.37.63.175 [1]. These are shared infrastructure (Mullvad, OVH, Datacamp) and may generate false positives, so correlate with CloudTrail timing.
Sigma Rule: Cross-Account AssumeRole from Qubole Account
title: Qubole Cross-Account AssumeRole Abuse
id: f3e1a7c2-9b4d-4e8f-a1c3-6d2e5f7b8a09
status: experimental
description: Detects AssumeRole calls originating from Qubole AWS account 805246085872 or using the quiam- session name prefix associated with the October 2026 supply chain campaign.
author: RedSheepSec
date: 2026/11/20
references:
- https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
logsource:
product: aws
service: cloudtrail
detection:
selection_account:
eventName: AssumeRole
requestParameters|contains: '805246085872'
selection_session:
eventName: AssumeRole
requestParameters|contains: 'quiam-'
condition: selection_account or selection_session
falsepositives:
- Legitimate Qubole platform operations from AWS infrastructure (validate source IP against known Qubole ranges)
level: high
tags:
- attack.initial_access
- attack.t1199
- attack.t1078.004
Dormant Role Identification
Use AWS IAM Access Analyzer or the IAM credential report to identify all cross-account roles where RoleLastUsed is older than 30 days or null. Cross-reference with trust policies that name external AWS accounts. Any role trusting an external account that has not been used in 30 days or more should be disabled or deleted.
Analysis
This campaign is a textbook supply-chain identity abuse: the attacker did not need to break any AWS service, exploit a vulnerability, or deploy malware. Every API call succeeded because the trust policies said it should. The attacker had the right account, the right IAM user, and the right External ID [12]. AWS's statement confirms this: its services "operated as designed" [6]. The failure is in how customers configured and maintained those trust relationships.
The attack surface was amplified by two factors. First, Qubole is an Idera-owned platform that many customers appear to have adopted and later abandoned without cleaning up IAM artifacts [4][5]. Second, the roles granted broad read permissions, allowing the attacker to access S3 data, secrets stores, and configuration data across multiple services [4]. Least-privilege principles were not applied to the vendor integration roles.
The credential types harvested extend beyond AWS. Entra ID service principal credentials and Okta OAuth client credentials were also stolen and validated [2][6]. This creates potential for lateral movement into Microsoft 365, Azure, and Okta-federated applications. The full scope of downstream compromise from these secondary credentials remains under investigation by Wiz and Mandiant [11].
The n0n group's claims of nearly one petabyte of exfiltrated data [7][9] and the naming of 59 organizations [8] should be treated with appropriate skepticism. Check Point's assessment that the financial institution claims likely trace back to a shared vendor, rather than direct intrusions, is consistent with the technical evidence [8]. The leaked data structure matches marketing and loyalty records, not payment systems [8].
AWS's recommendation to customers, as relayed through its statement to The Register, includes regularly reviewing cross-account IAM roles, removing access for integrations no longer in use, granting least-privilege permissions, and adding conditions such as source IP restrictions to role trust policies [6]. These are standard practices, but this incident demonstrates that many organizations do not implement them, particularly for vendor integrations that were configured years ago.
Red Sheep Assessment
Assessment (Moderate-High Confidence): The Qubole campaign pattern, compromising a trusted vendor's IAM credentials to inherit standing access across dozens of customer environments, is directly replicable against any SaaS or data platform that uses cross-account IAM roles. The technique requires no exploit, no malware, and no novel tooling. The attacker's operational sequence (enumerate, harvest secrets, validate credentials, exfiltrate) was identical across every victim, suggesting automation or at minimum a scripted runbook. The use of TruffleHog for credential validation and the quiam- session name convention indicate operational discipline but not exceptional technical skill.
The secondary credential harvest (Entra ID, Okta) is the more concerning vector for downstream impact. Organizations that stored Entra ID service principal secrets or Okta client credentials in AWS Secrets Manager, SSM Parameter Store, or Lambda environment variables and had a Qubole trust role in place should assume those credentials are compromised and rotate them, regardless of whether they can confirm direct evidence of access in their own CloudTrail logs.
This pattern extends beyond this single vendor. CISA's own 2026 GitHub credential leak [14], the Barracuda vendor-key incident described in the same week's reporting [5], and this Qubole campaign all share the same root cause: credentials granted to a third party for a legitimate purpose, left active after the purpose ended, and eventually used by an unauthorized actor. The industry has a systematic failure in credential lifecycle management for third-party integrations, and attackers are exploiting that failure at scale.
One alternative interpretation: n0n may be overstating its haul to maximize extortion pressure. The 59-organization leak site listing could include names derived from marketing data exports (campaign directories with brand names) rather than confirmed individual intrusions into each named company. The consistent file structure across brands [8] supports this reading. Defenders at named organizations should focus on what their own logs show rather than reacting solely to the leak site claims.
Defender's Checklist
- ▢[ ] Audit all cross-account IAM trust relationships in every AWS account for roles trusting external accounts. Prioritize roles where
RoleLastUsedis null or older than 30 days. Delete or disable any role trusting account805246085872that is not actively required. Useaws iam get-account-authorization-details --filter Roleand parse trust policies.
- ▢[ ] Search CloudTrail for Qubole-linked activity from October 1, 2026 forward, using the documented account ID (
805246085872), IAM principal IDs (AIDA3W7DIFLYOLTK63SFS,AIDAJYRRH5Y3A7AAJM2HW), session name prefix (quiam-), and the four documented source IPs [1][5].
- ▢[ ] Rotate all credentials stored in Secrets Manager, SSM Parameter Store, and Lambda environment variables in any AWS account that had an active Qubole trust role, regardless of whether direct compromise has been confirmed. This includes AWS access keys, Entra ID service principal secrets, and Okta client credentials [2][6].
- ▢[ ] Add
aws:SourceIporaws:SourceVpcconditions to trust policies for any remaining cross-account vendor roles. External ID alone does not prevent abuse when the caller already holds valid IAM credentials [6][12].
- ▢[ ] Establish a recurring review cadence (quarterly minimum) for all third-party IAM integrations. Require vendor re-validation for any trust relationship older than 12 months. Document the business owner for each cross-account role so that abandoned integrations have an accountable party.
References
[1] https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
[2] https://www.csoonline.com/article/4246780/qubole-supply-chain-attack-exposed-aws-entra-id-and-okta-credentials.html
[3] https://www.bleepingcomputer.com/news/security/hackers-exploit-trusted-aws-access-to-breach-multiple-cloud-environments/
[4] https://www.securityweek.com/over-a-dozen-organizations-hit-in-qubole-supply-chain-attack/
[5] https://sherlockforensics.com/blog/2026-11-20-weekly-security-roundup.html
[6] https://www.rescana.com/post/qubole-aws-cross-account-trust-abuse-prepare-now
[7] https://www.bleepingcomputer.com/news/security/vendor-data-breach-exposes-chase-visa-mastercard-customer-info/
[8] https://cyberint.com/blog/threat-intelligence/threat-actor-investigation-n0n/
[9] https://cyberinsider.com/jpmorgan-chase-vendor-cyberattack-customer-data/
[10] https://www.crowdfundinsider.com/2026/11/323252-hackers-exploit-quboles-trusted-aws-access-amplifying-risks-to-digital-infrastructure-and-fintech-platforms/
[11] https://aviatrix.ai/threat-research-center/hackers-exploit-trusted-aws-access-breach-multiple-cloud-environments-2026/
[12] https://safeguard.sh/resources/blog/qubole-aws-cross-account-role-abuse-november-2026
[13] https://itsec.asia/blog/your-cloud-vendor-has-access-what-happens-when-its-credentials-are-compromised
[14] https://axis-intelligence.com/cisa-github-data-leak-full-technical-review/
Visual Intelligence
Timeline (14 events)
Entity Graph (12 entities, 10 relationships)
Diamond Model
---
Hunt Guide: Qubole Cross-Account IAM Credential Compromise (n0n Group)
Attribution: Rules adapted, ported, or quoted from a public source retain that source's author (e.g. SigmaHQ / Florian Roth, Elastic, Emerging Threats, Abuse.ch, or the cited vendor/researcher) and carry an attribution credit. Rules without a credit were authored in-house by RedSheepSec. If you reuse a rule, preserve its stated attribution.
Hypothesis: If the n0n threat actor or a similarly positioned adversary has abused Qubole cross-account IAM trust relationships to access our AWS environments, we expect to observe AssumeRole events sourced from AWS account 805246085872 with session name prefixes beginning with quiam-, originating from non-AWS IP addresses (particularly 45.83.220.211, 185.209.199.73, 57.129.54.120, 89.37.63.175), followed by rapid enumeration of S3 buckets, IAM identities, and secrets stores in AWS CloudTrail logs.
Intelligence Summary: Between October 14 and 26, 2026, a threat actor used compromised IAM credentials from Qubole's AWS account (805246085872) to assume cross-account IAM roles in dozens of downstream customer AWS environments. The attacker followed a repeatable playbook: enumerate cloud resources, harvest secrets from AWS Secrets Manager, SSM Parameter Store, and Lambda environment variables, validate stolen credentials with TruffleHog, and bulk-download S3 data. An actor calling themselves n0n claimed responsibility on November 11, 2026, and named 59 organizations on its leak site; Check Point assesses with moderate confidence that claims against major financial institutions trace to shared data vendors rather than direct intrusions.
Confidence: High | Priority: Critical
Scope
- Networks: All AWS accounts within the organization, with priority on accounts that have or previously had Qubole (Idera) integrations. Secondary scope includes Azure/Entra ID and Okta tenants if credentials were stored in AWS secrets stores.
- Timeframe: October 1, 2026 through present. Primary activity window: October 14 to October 26, 2026. Extend to present to capture any delayed exploitation of harvested secondary credentials (Entra ID, Okta).
- Priority Systems: AWS accounts with cross-account trust roles referencing account 805246085872. Secrets Manager vaults and SSM Parameter Stores containing Entra ID service principal secrets or Okta OAuth client credentials. S3 buckets containing PII, PHI, or financial data.
MITRE ATT&CK Techniques
T1199: Trusted Relationship (Initial Access) [P1]
The attacker used compromised Qubole IAM credentials to exercise pre-existing cross-account IAM trust relationships that customers had configured for Qubole integrations. Many of these roles were dormant, meaning the trust policy was still active but the customer had stopped using Qubole.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName=AssumeRole
| where like(requestParameters, "%805246085872%") OR like(requestParameters, "%quiam-%")
| table _time, recipientAccountId, requestParameters, sourceIPAddress, userIdentity.arn, userIdentity.accountId
| sort - _time
Elastic KQL:
event.action:"AssumeRole" AND (aws.cloudtrail.request_parameters:"805246085872" OR aws.cloudtrail.request_parameters:"quiam-")
Sigma Rule:
title: Qubole Cross-Account AssumeRole Abuse
id: f3e1a7c2-9b4d-4e8f-a1c3-6d2e5f7b8a09
status: experimental
description: Detects AssumeRole calls originating from Qubole AWS account 805246085872 or using the quiam- session name prefix associated with the October 2026 supply chain campaign.
author: RedSheepSec
date: 2026/11/20
references:
- https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
logsource:
product: aws
service: cloudtrail
detection:
selection_account:
eventName: AssumeRole
requestParameters|contains: '805246085872'
selection_session:
eventName: AssumeRole
requestParameters|contains: 'quiam-'
condition: selection_account or selection_session
falsepositives:
- Legitimate Qubole platform operations from AWS infrastructure; validate source IP against known Qubole ranges
level: high
tags:
- attack.initial_access
- attack.t1199
- attack.t1078.004
Any AssumeRole from account 805246085872 to your environment after October 1, 2026 warrants immediate investigation. If Qubole is not an active vendor, all such events are anomalous. The quiam- session name prefix was used consistently by the attacker and is a high-fidelity indicator.
T1078.004: Valid Accounts: Cloud Accounts (Initial Access) [P1]
Two legitimate Qubole IAM users (AIDA3W7DIFLYOLTK63SFS and AIDAJYRRH5Y3A7AAJM2HW) were used by the attacker to authenticate to AWS and assume cross-account roles. The first principal was observed only in threat actor activity; the second also carries legitimate Qubole traffic.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail"
| where like(userIdentity.principalId, "%AIDA3W7DIFLYOLTK63SFS%") OR like(userIdentity.principalId, "%AIDAJYRRH5Y3A7AAJM2HW%")
| table _time, eventName, sourceIPAddress, recipientAccountId, requestParameters, userIdentity.arn
| sort - _time
Elastic KQL:
aws.cloudtrail.user_identity.principal_id:("AIDA3W7DIFLYOLTK63SFS" OR "AIDAJYRRH5Y3A7AAJM2HW")
Sigma Rule:
title: Qubole Compromised IAM Principal Activity
id: a8c2d1e4-5f6b-47a9-b3c8-9d1e2f4a5b67
status: experimental
description: Detects CloudTrail activity from IAM principal IDs associated with the compromised Qubole account used in the October 2026 campaign.
author: RedSheepSec
date: 2026/11/20
references:
- https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
logsource:
product: aws
service: cloudtrail
detection:
selection:
userIdentity.principalId|contains:
- 'AIDA3W7DIFLYOLTK63SFS'
- 'AIDAJYRRH5Y3A7AAJM2HW'
condition: selection
falsepositives:
- AIDAJYRRH5Y3A7AAJM2HW may appear in legitimate Qubole operations; correlate with source IP to distinguish
level: critical
tags:
- attack.initial_access
- attack.t1078.004
AIDA3W7DIFLYOLTK63SFS was observed only in threat actor activity per Wiz. Any event from this principal is high confidence malicious. AIDAJYRRH5Y3A7AAJM2HW also carries legitimate Qubole traffic, so correlate with source IP.
T1550.001: Use Alternate Authentication Material: Application Access Token (Defense Evasion) [P1]
Temporary STS credentials obtained via AssumeRole were used to operate within victim environments. The attacker received short-lived tokens scoped to the assumed role's permissions.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName=AssumeRole
| where like(requestParameters, "%805246085872%")
| eval session_name=mvindex(split(requestParameters, "roleSessionName"), 1)
| where like(session_name, "%quiam-%")
| stats count by sourceIPAddress, recipientAccountId, userIdentity.arn
| sort - count
Elastic KQL:
event.action:"AssumeRole" AND aws.cloudtrail.request_parameters:*805246085872* AND aws.cloudtrail.request_parameters:*quiam-*
Look for the temporary credentials being used from non-AWS source IPs, especially Mullvad VPN or OVH address ranges.
T1526: Cloud Service Discovery (Discovery) [P2]
After assuming each role, the attacker enumerated S3 buckets (ListBuckets, ListObjects), EC2 instances (DescribeInstances), security groups (DescribeSecurityGroups), and databases in the victim environment.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName IN ("ListBuckets", "ListObjects", "DescribeInstances", "DescribeSecurityGroups", "DescribeDBInstances")
| where like(userIdentity.arn, "%quiam-%") OR like(userIdentity.accountId, "%805246085872%")
| stats count by eventName, sourceIPAddress, userIdentity.arn, recipientAccountId
| sort - count
Elastic KQL:
event.action:("ListBuckets" OR "ListObjects" OR "DescribeInstances" OR "DescribeSecurityGroups" OR "DescribeDBInstances") AND (aws.cloudtrail.user_identity.arn:*quiam-* OR aws.cloudtrail.user_identity.account_id:"805246085872")
Sigma Rule:
title: Post-AssumeRole Cloud Service Enumeration Burst
id: c7d8e9f0-1a2b-3c4d-5e6f-7a8b9c0d1e2f
status: experimental
description: Detects a burst of cloud resource enumeration API calls following a cross-account AssumeRole, consistent with the Qubole campaign post-compromise playbook.
author: RedSheepSec
date: 2026/11/20
references:
- https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
logsource:
product: aws
service: cloudtrail
detection:
selection:
eventName:
- ListBuckets
- ListObjects
- DescribeInstances
- DescribeSecurityGroups
- ListRoles
- ListUsers
- GetAccountAuthorizationDetails
userIdentity.arn|contains: 'quiam-'
condition: selection
falsepositives:
- Automated cloud inventory tools may produce similar patterns; validate session name and source IP
level: high
tags:
- attack.discovery
- attack.t1526
The enumeration pattern is distinctive when it follows an AssumeRole from an external account. Focus on sessions with quiam- prefix. Legitimate cloud inventory tools may trigger this; use the session name and source IP for disambiguation.
T1087.004: Account Discovery: Cloud Account (Discovery) [P2]
The attacker enumerated IAM identities within victim environments using ListRoles, ListUsers, and GetAccountAuthorizationDetails.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName IN ("ListRoles", "ListUsers", "GetAccountAuthorizationDetails")
| where like(userIdentity.arn, "%quiam-%")
| table _time, eventName, sourceIPAddress, userIdentity.arn, recipientAccountId
| sort - _time
Elastic KQL:
event.action:("ListRoles" OR "ListUsers" OR "GetAccountAuthorizationDetails") AND aws.cloudtrail.user_identity.arn:*quiam-*
GetAccountAuthorizationDetails returns a comprehensive dump of all IAM policies, users, groups, and roles. This call from a cross-account assumed role is high signal.
T1552.007: Unsecured Credentials: Cloud Secrets Management Stores (Credential Access) [P1]
The attacker harvested secrets from AWS Secrets Manager (GetSecretValue), SSM Parameter Store (GetParameter, GetParameters, GetParametersByPath), and Lambda environment variables. Sentra documented approximately 100 passwords and keys read in 7 minutes in one session.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName IN ("GetSecretValue", "GetParameter", "GetParameters", "GetParametersByPath", "GetFunction", "GetFunctionConfiguration")
| where like(userIdentity.arn, "%quiam-%") OR sourceIPAddress IN ("45.83.220.211", "185.209.199.73", "57.129.54.120", "89.37.63.175")
| stats count dc(requestParameters) as unique_secrets by sourceIPAddress, userIdentity.arn, recipientAccountId
| where count > 10
| sort - count
Elastic KQL:
event.action:("GetSecretValue" OR "GetParameter" OR "GetParameters" OR "GetParametersByPath" OR "GetFunction" OR "GetFunctionConfiguration") AND (aws.cloudtrail.user_identity.arn:*quiam-* OR source.ip:("45.83.220.211" OR "185.209.199.73" OR "57.129.54.120" OR "89.37.63.175"))
Sigma Rule:
title: Rapid Secrets Harvesting from AWS Secrets Manager and SSM
id: d9e0f1a2-3b4c-5d6e-7f8a-9b0c1d2e3f4a
status: experimental
description: Detects a burst of GetSecretValue or GetParameter calls within a short window, consistent with the Qubole campaign secret harvesting playbook where approximately 100 credentials were read in 7 minutes.
author: RedSheepSec
date: 2026/11/20
references:
- https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
- https://sherlockforensics.com/blog/2026-11-20-weekly-security-roundup.html
logsource:
product: aws
service: cloudtrail
detection:
selection:
eventName:
- GetSecretValue
- GetParameter
- GetParameters
- GetParametersByPath
condition: selection
falsepositives:
- Application startup that reads multiple configuration parameters; baseline normal read patterns per role
level: high
tags:
- attack.credential_access
- attack.t1552.007
The 100-secrets-in-7-minutes pattern is distinctive. Threshold tuning required: baseline normal GetSecretValue frequency per role. Cross-account assumed roles reading secrets they have never accessed before are high priority.
T1552: Unsecured Credentials (Credential Access) [P2]
Secrets retrieved from Lambda environment variables, which store credentials in runtime configuration. The attacker used GetFunction and GetFunctionConfiguration to access these values.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName IN ("GetFunction", "GetFunctionConfiguration")
| where like(userIdentity.arn, "%quiam-%") OR like(userIdentity.accountId, "%805246085872%")
| table _time, eventName, sourceIPAddress, userIdentity.arn, requestParameters, recipientAccountId
| sort - _time
Elastic KQL:
event.action:("GetFunction" OR "GetFunctionConfiguration") AND (aws.cloudtrail.user_identity.arn:*quiam-* OR aws.cloudtrail.user_identity.account_id:"805246085872")
GetFunctionConfiguration returns environment variables which may contain hardcoded credentials, API keys, and connection strings. Any cross-account caller reading Lambda configurations warrants review.
T1530: Data from Cloud Storage (Collection) [P1]
The attacker downloaded files from accessible S3 buckets, in some cases exfiltrating entire buckets. Sentra confirmed millions of files stolen from at least one organization.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName IN ("GetObject", "ListObjects", "ListObjectsV2")
| where like(userIdentity.arn, "%quiam-%") OR sourceIPAddress IN ("45.83.220.211", "185.209.199.73", "57.129.54.120", "89.37.63.175")
| stats count sum(bytes) as total_bytes dc(requestParameters) as unique_objects by sourceIPAddress, userIdentity.arn, recipientAccountId
| where count > 100
| sort - total_bytes
Elastic KQL:
event.action:("GetObject" OR "ListObjects" OR "ListObjectsV2") AND (aws.cloudtrail.user_identity.arn:*quiam-* OR source.ip:("45.83.220.211" OR "185.209.199.73" OR "57.129.54.120" OR "89.37.63.175"))
S3 data events must be enabled in CloudTrail for GetObject visibility. Many environments only log management events by default. If data events are not enabled, look for ListBuckets and ListObjects as proxies for exfiltration staging.
T1090.003: Proxy: Multi-hop Proxy (Command and Control) [P1]
The attacker routed all observed activity through Mullvad VPN exits (Sweden), an OVH VPS (Germany), and a Datacamp-hosted server (Sweden). No activity originated from Qubole's own network.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" sourceIPAddress IN ("45.83.220.211", "185.209.199.73", "57.129.54.120", "89.37.63.175")
| stats count values(eventName) as events values(recipientAccountId) as accounts by sourceIPAddress
| sort - count
Elastic KQL:
source.ip:("45.83.220.211" OR "185.209.199.73" OR "57.129.54.120" OR "89.37.63.175")
These IPs are shared infrastructure (Mullvad VPN, OVH, Datacamp) and may generate false positives from legitimate VPN users. Correlate with CloudTrail event types and timing from the October 14-26, 2026 window.
T1078.004: Valid Accounts: Cloud Accounts (Defense Evasion) [P2]
The attacker used TruffleHog to call sts:GetCallerIdentity with discovered IAM access keys in rapid bursts, validating harvested credentials. Stolen credential types included AWS access keys, Entra ID service principal credentials, and Okta OAuth client credentials.
Splunk SPL:
index=cloud-aws sourcetype="aws:cloudtrail" eventName="GetCallerIdentity"
| bin _time span=60s
| stats count dc(userIdentity.accessKeyId) as unique_keys by _time, sourceIPAddress
| where count > 5 AND unique_keys > 3
| sort - count
Elastic KQL:
event.action:"GetCallerIdentity" AND source.ip:("45.83.220.211" OR "185.209.199.73" OR "57.129.54.120" OR "89.37.63.175")
Sigma Rule:
title: Rapid GetCallerIdentity Credential Validation Burst
id: e1f2a3b4-5c6d-7e8f-9a0b-1c2d3e4f5a6b
status: experimental
description: Detects rapid bursts of sts GetCallerIdentity calls from a single source IP using multiple distinct access keys, indicating programmatic credential validation consistent with TruffleHog usage in the Qubole campaign.
author: RedSheepSec
date: 2026/11/20
references:
- https://www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild
logsource:
product: aws
service: cloudtrail
detection:
selection:
eventName: GetCallerIdentity
condition: selection
falsepositives:
- CI/CD pipelines or automation that validate credentials at startup
level: medium
tags:
- attack.defense_evasion
- attack.t1078.004
GetCallerIdentity is benign in isolation. The behavioral indicator is multiple calls from the same source IP using distinct access key IDs within seconds. The Sigma rule requires SIEM-side aggregation to detect the burst pattern; the Splunk query implements this with bin and stats.
Indicators of Compromise
| Type | Value | Context | |
|---|---|---|---|
| ip | 45.83.220.211 |
Mullvad VPN exit (Sweden) used for AssumeRole calls against customer environments from Qubole account \ | AbuseIPDB confidence 100% (191 reports, SE) |
| ip | 185.209.199.73 |
Second Mullvad VPN exit (Sweden) used to assume integration roles in customer AWS environments \ | AbuseIPDB confidence 15% (7 reports, SE) |
| ip | 57.129.54.120 |
OVH VPS (Germany) used by the threat actor during the campaign \ | AbuseIPDB confidence 0% (0 reports, DE) |
| ip | 89.37.63.175 |
Datacamp-hosted IP (Sweden) used to assume customer roles during the campaign \ | AbuseIPDB confidence 0% (41 reports, SE) |
| filename | AIDA3W7DIFLYOLTK63SFS |
IAM Principal ID in Qubole AWS account 805246085872; observed only in threat actor activity in Wiz-monitored environments | |
| filename | AIDAJYRRH5Y3A7AAJM2HW |
Second IAM Principal ID in Qubole AWS account; used from October 18 onward, also carries legitimate Qubole traffic | |
| filename | 805246085872 |
Qubole AWS Account ID used as the source account for all observed malicious cross-account role assumptions | |
| filename | quiam- |
Role session name prefix used by the attacker across all observed assumed-role sessions; reliable CloudTrail hunting indicator |
IOC Sweep Queries (Splunk):
index=cloud-aws sourcetype="aws:cloudtrail" sourceIPAddress="45.83.220.211"
| table _time, eventName, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" sourceIPAddress="185.209.199.73"
| table _time, eventName, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" sourceIPAddress="57.129.54.120"
| table _time, eventName, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" sourceIPAddress="89.37.63.175"
| table _time, eventName, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" "AIDA3W7DIFLYOLTK63SFS"
| table _time, eventName, sourceIPAddress, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" "AIDAJYRRH5Y3A7AAJM2HW"
| table _time, eventName, sourceIPAddress, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" "805246085872"
| table _time, eventName, sourceIPAddress, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
index=cloud-aws sourcetype="aws:cloudtrail" "quiam-"
| table _time, eventName, sourceIPAddress, recipientAccountId, userIdentity.arn, requestParameters
| sort - _time
Suricata Rules
SID 2026111: Detects outbound connection to Mullvad VPN IP 45.83.220.211 associated with Qubole cross-account attack
alert ip any any -> 45.83.220.211 any (msg:"HUNT - Qubole Campaign Mullvad VPN Exit 45.83.220.211"; reference:url,www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild; classtype:trojan-activity; sid:2026111; rev:1;)
SID 2026112: Detects outbound connection to Mullvad VPN IP 185.209.199.73 associated with Qubole cross-account attack
alert ip any any -> 185.209.199.73 any (msg:"HUNT - Qubole Campaign Mullvad VPN Exit 185.209.199.73"; reference:url,www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild; classtype:trojan-activity; sid:2026112; rev:1;)
SID 2026113: Detects outbound connection to OVH VPS 57.129.54.120 associated with Qubole cross-account attack
alert ip any any -> 57.129.54.120 any (msg:"HUNT - Qubole Campaign OVH VPS 57.129.54.120"; reference:url,www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild; classtype:trojan-activity; sid:2026113; rev:1;)
SID 2026114: Detects outbound connection to Datacamp IP 89.37.63.175 associated with Qubole cross-account attack
alert ip any any -> 89.37.63.175 any (msg:"HUNT - Qubole Campaign Datacamp IP 89.37.63.175"; reference:url,www.wiz.io/blog/qubole-cross-account-access-abused-in-the-wild; classtype:trojan-activity; sid:2026114; rev:1;)
Data Source Requirements
| Source | Required For | Notes |
|---|---|---|
| AWS CloudTrail (Management Events) | T1199, T1078.004, T1550.001, T1087.004, T1526, T1552.007, T1552, T1090.003 | CloudTrail must be enabled in all AWS accounts and forwarded to the SIEM. Verify that multi-region trails are active. The index cloud-aws with sourcetype aws:cloudtrail is the target. Confirm logs cover October 1, 2026 forward. |
| AWS CloudTrail (S3 Data Events) | T1530 | S3 data events (GetObject, PutObject) must be explicitly enabled in CloudTrail configuration. Many environments log only management events. Without data events, S3 exfiltration visibility is limited to ListBuckets and ListObjects. |
| Firewall/Proxy Logs | T1090.003 | Firewall logs (index=firewall-pan with pan:traffic:aggregated or pan:threat) should be searched for connections to the four attacker IPs. Correlate timing with CloudTrail events. |
| Network Metadata (Zeek/Corelight) | T1090.003 | Corelight conn logs (index=corelight, sourcetype corelight_conn) can be searched for connections to attacker IPs on the perimeter. |
| IAM Access Analyzer / IAM Credential Report | T1199 | Required for identifying dormant cross-account trust roles. Use aws iam get-account-authorization-details to enumerate all roles and their trust policies. Not a SIEM data source but a critical AWS-native audit. |
Recommendations
- Audit all AWS accounts for cross-account IAM trust roles referencing external accounts. Delete or disable any role trusting account 805246085872 that is not actively required. Use aws iam get-account-authorization-details to enumerate roles and parse trust policies.
- Search CloudTrail from October 1, 2026 forward using the documented IOCs: account ID 805246085872, IAM principal IDs AIDA3W7DIFLYOLTK63SFS and AIDAJYRRH5Y3A7AAJM2HW, session name prefix quiam-, and the four attacker source IPs.
- Rotate all credentials stored in AWS Secrets Manager, SSM Parameter Store, and Lambda environment variables in any account that had an active Qubole trust role. This includes AWS access keys, Entra ID service principal secrets, and Okta OAuth client credentials.
- Add aws:SourceIp or aws:SourceVpc conditions to trust policies for all remaining cross-account vendor roles. External ID alone does not prevent abuse when the caller holds valid IAM credentials from the trusted account.
- Deploy the Splunk and Sigma detection analytics from this report across all SIEM instances ingesting AWS CloudTrail data. Enable S3 data events in CloudTrail for accounts containing sensitive data to gain GetObject visibility.
- Establish a quarterly review cadence for all third-party IAM integrations. Require vendor re-validation for trust relationships older than 12 months. Assign a business owner for each cross-account role.
- Block or alert on the four documented attacker IPs (45.83.220.211, 185.209.199.73, 57.129.54.120, 89.37.63.175) at the perimeter, with the caveat that these are shared VPN and hosting infrastructure and may carry legitimate traffic.
Sources
- Wiz Blog: Qubole Cross-Account Access Abused in the Wild
- CSO Online: Qubole Supply Chain Attack Exposed AWS, Entra ID, and Okta Credentials
- BleepingComputer: Hackers Exploit Trusted AWS Access to Breach Multiple Cloud Environments
- SecurityWeek: Over a Dozen Organizations Hit in Qubole Supply Chain Attack
- Sherlock Forensics: Weekly Security Roundup 2026-11-20
- Rescana: Qubole AWS Cross-Account Trust Abuse
- BleepingComputer: Vendor Data Breach Exposes Chase, Visa, Mastercard Customer Info
- Cyberint: Threat Actor Investigation - n0n
- CyberInsider: JPMorgan Chase Vendor Cyberattack Customer Data
- Crowdfund Insider: Hackers Exploit Qubole's Trusted AWS Access
- Aviatrix: Hackers Exploit Trusted AWS Access Breach Multiple Cloud Environments
- Safeguard: Qubole AWS Cross-Account Role Abuse November 2026
- ITSEC Asia: Your Cloud Vendor Has Access - What Happens When Its Credentials Are Compromised
- Axis Intelligence: CISA GitHub Data Leak Full Technical Review